What drives inflation and how? Evidence from additive mixed models selected by cAIC

Abstract

We analyze the forces that explain inflation using a panel of 122 countries from 1997 to 2015 with 37 regressors. Ninety-eight models motivated by economic theory are compared to a boosting algorithm, non-linearities and structural breaks are considered. We show that the typical estimation methods are likely to lead to fallacious policy conclusions, which motivates the use of a new approach that we propose in this paper. The boosting algorithm outperforms theory-based models. Furthermore, we extend the current software implementation of conditional Akaike Information Criteria for additive mixed models with observation weights. We present a novel two-step selection process suitable for a wide range of applications that enables to empirically compare theory- and data-driven models with varying data availability.